Twitch Put Its New Amazon AI-Training Switch in the Off Position Users Have to Find

Twitch has added a privacy control that lets creators refuse the use of their channel content for generative-AI training across Amazon. The switch is not a request to join the programme. It is an exit: creators who do nothing remain available for training.

The announcement arrived through Twitch Support on X rather than a full Twitch Blog post. Twitch's account-settings FAQ says permitted content may be used for future improvements to generative models. Its example says audio could refine speech-to-text systems used for captions on Twitch and elsewhere at Amazon.

The boundary matters. Turning off ‘Training for Generative AI’ does not remove a channel from every AI or machine-learning system on Twitch. The FAQ separates generative-model training from features such as automated captions and other services that analyse content without retaining it to teach models to create new output.

Creators can find the switch under Settings, then Security and Privacy. The release has drawn criticism because consent is assumed until the creator finds and changes it. Twitch Support has also tried to distinguish between Twitch and Amazon by saying Twitch itself is not training generative models on streamer content. The practical effect described by its own announcement remains that channel content can be used for model training across Amazon unless the creator opts out.

The setting does not answer every question about earlier material, licensed games visible in streams or what content has already been processed. Twitch has not publicly established that any particular creator's archive has already trained a model. What it does establish is a new choice that is easy to miss. Whatever creators decide, the first step is simple: check the setting rather than assuming the announcement changed nothing.